Method configure

src/gpu/cl/kernels/ClGemmLowpReductionKernel.cpp:107–152

}

void ClGemmLowpMatrixAReductionKernel::configure(const CLCompileContext            &compile_context,
                                                 const ITensorInfo                 *mtx_a,
                                                 ITensorInfo                       *vector_sum_row,
                                                 const GEMMLowpReductionKernelInfo &info)
{
    // Perform validate step
    ARM_COMPUTE_ERROR_ON_NULLPTR(mtx_a, vector_sum_row);
    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_matrix_a_reduction(mtx_a, vector_sum_row));

    // Output auto initialization if not yet initialized
    auto_init_if_empty(*vector_sum_row, TensorShape(mtx_a->dimension(1)), 1, DataType::S32);

    auto padding_info = get_padding_info({mtx_a, vector_sum_row});

    // Set the arguments to pass at compile time
    CLBuildOptions build_opts;
    build_opts.add_option("-DCOLS_A=" + support::cpp11::to_string(mtx_a->dimension(0)));
    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(mtx_a->data_type()));
    build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_dot8_acc_type_from_data_type(mtx_a->data_type()));
    build_opts.add_option_if(info.mul_by_scalar, "-DSCALAR=" + support::cpp11::to_string(info.scalar));

    const bool is_dot8_supported = dot8_supported(CLKernelLibrary::get().get_device());

    std::string kernel_name = "gemmlowp_matrix_a_reduction" + std::string(is_dot8_supported ? "_dot8" : "");

    // A macro guard to compile ONLY the kernel of interest
    build_opts.add_option("-D" + upper_string(kernel_name));

    // Create kernel
    _kernel = create_kernel(compile_context, kernel_name, build_opts.options());

    // Configure kernel window
    // This kernel does not need padding
    Window win = calculate_max_window(*vector_sum_row, Steps());
    ICLKernel::configure_internal(win);

    _config_id = kernel_name;
    _config_id += "_";
    _config_id += support::cpp11::to_string(mtx_a->dimension(0));
    _config_id += "_";
    _config_id += support::cpp11::to_string(mtx_a->dimension(1));
    _config_id += "_";
    _config_id += support::cpp11::to_string(mtx_a->dimension(2));

    ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
}

Status ClGemmLowpMatrixAReductionKernel::validate(const ITensorInfo                 *mtx_a,
                                                  const ITensorInfo                 *vector_sum_row,